Cost savings and usage time
Fewer tokens. More work. Same plan.
Token Saver reduces token consumption across 508 production plans. Pay-per-token buyers save money. Subscription buyers get more work from the same allowance — without changing their plan or paying more.
Pay-per-token buyers
The same work for 80% less
Step-level measurement across 167 plans and 669 measured steps — the same account and window as the figures above, summed before the dashboard's step filters (which is why the dollar saving here is larger than the reported figure).
Step-level measurement across 167 plans and 669 measured steps.
Figures are pay-per-token equivalents. On a subscription, Token Saver saves usage against your allowance rather than per-token dollars — you get more work from the same plan.
Subscription buyers
Token savings translate to more work per allowance
On a subscription, Token Saver savings show up as more use from the same plan allowance. Across 508 production plans, the same work took 44.8% fewer tokens — roughly 1.8× the work from one allowance.
Token ratio spread across all measured plans
Measured on 508 production plans: 1,361,307 baseline tokens against 751,419 optimised, measured per plan in production.
How a 5-hour work window extends under each scenario
| Percentile | Ratio | Work time | Case |
|---|---|---|---|
| p10 | 1.5× | 7.5h | Minimum case — still 50% more work from one allowance |
| p50 | 1.73× | 8.7h | Median case — 73% more work from one allowance |
| mean | 1.81× | 9.1h | Average across all plans — 81% more work, or weekly cap does not run out |
| p90 | 2.53× | 12.6h | Upper case — more than double the work from one allowance |
Weekly caps
At the baseline, a developer's weekly cap runs out on Thursday. Under token savings:
- At p10 or p50: The cap lasts until Saturday instead
- At the mean (1.81×): The weekly cap does not run out — work extends past Friday evening
- The published "Friday evening" line is a deliberate underpromise at p50.
Per-plan rate limits
A limit that used to bind after about five hours of heavy use stops binding inside a working day. At p50 (1.73×), a developer's daily work extends from a half day to roughly a full working day before hitting the rate cap.
When this applies
Token-metered hosts (savings extend usage):
- • Claude Pro / Max (token-metered)
- • ChatGPT / Codex with 5-hour windows + weekly caps (token-metered)
Request-metered hosts (not extended by token savings):
- • GitHub Copilot premium (request-metered, not token-metered)
- • Cursor credits (request-metered, not token-metered)
- • Wall-clock per-task and per-minute rate limits (unchanged by token savings)
ChatGPT / Codex with 5-hour windows + weekly caps (token-metered)
Full study
Read the complete Token Saver subscription whitepaper: methodology, baseline assumptions, usage scenarios for different host plans, and how token savings translate to subscription value across different adoption profiles.
Sample: one real production account — a power user running ~29 plans a day. Reported, not projected; not an average-user claim. Collection since 2026-08-18; 15.5 active days inside the 30-day window.